GPT-Live-1 Comes to the API, Listening and Speaking at Once

OpenAI made GPT-Live-1 available in the API on 10 September 2026. The model was introduced first in ChatGPT; this release opens it to developers building voice-enabled applications.

The distinguishing property OpenAI names is full duplex — the model listens and speaks at the same time, rather than taking turns. Its interruption handling is described as coming from a single model that reasons over incoming and outgoing audio together, which OpenAI contrasts with "the latency and brittle handoffs of chained STT-LLM-TTS architectures".

OpenAI cites one early evaluation by name: the language-learning company Speak found that GPT-Live-1 gave learners more time to think before the tutor responded, cutting interruptions by almost 80% against their previous turn-based system. That figure is Speak's, reported by OpenAI, and covers one product rather than a benchmark.

As in ChatGPT and Codex, the model can delegate deeper reasoning and actions to the models and tools it is paired with, rather than performing them itself.